Mercurial > repos > xuebing > sharplabtool
view tools/maf/interval2maf.py @ 0:9071e359b9a3
Uploaded
author | xuebing |
---|---|
date | Fri, 09 Mar 2012 19:37:19 -0500 |
parents | |
children |
line wrap: on
line source
#!/usr/bin/env python """ Reads a list of intervals and a maf. Produces a new maf containing the blocks or parts of blocks in the original that overlapped the intervals. If a MAF file, not UID, is provided the MAF file is indexed before being processed. NOTE: If two intervals overlap the same block it will be written twice. usage: %prog maf_file [options] -d, --dbkey=d: Database key, ie hg17 -c, --chromCol=c: Column of Chr -s, --startCol=s: Column of Start -e, --endCol=e: Column of End -S, --strandCol=S: Column of Strand -t, --mafType=t: Type of MAF source to use -m, --mafFile=m: Path of source MAF file, if not using cached version -I, --mafIndex=I: Path of precomputed source MAF file index, if not using cached version -i, --interval_file=i: Input interval file -o, --output_file=o: Output MAF file -p, --species=p: Species to include in output -P, --split_blocks_by_species=P: Split blocks by species -r, --remove_all_gap_columns=r: Remove all Gap columns -l, --indexLocation=l: Override default maf_index.loc file -z, --mafIndexFile=z: Directory of local maf index file ( maf_index.loc or maf_pairwise.loc ) """ #Dan Blankenberg from galaxy import eggs import pkg_resources; pkg_resources.require( "bx-python" ) from bx.cookbook import doc_optparse import bx.align.maf import bx.intervals.io from galaxy.tools.util import maf_utilities import sys assert sys.version_info[:2] >= ( 2, 4 ) def __main__(): index = index_filename = None mincols = 0 #Parse Command Line options, args = doc_optparse.parse( __doc__ ) if options.dbkey: dbkey = options.dbkey else: dbkey = None if dbkey in [None, "?"]: maf_utilities.tool_fail( "You must specify a proper build in order to extract alignments. You can specify your genome build by clicking on the pencil icon associated with your interval file." ) species = maf_utilities.parse_species_option( options.species ) if options.chromCol: chromCol = int( options.chromCol ) - 1 else: maf_utilities.tool_fail( "Chromosome column not set, click the pencil icon in the history item to set the metadata attributes." ) if options.startCol: startCol = int( options.startCol ) - 1 else: maf_utilities.tool_fail( "Start column not set, click the pencil icon in the history item to set the metadata attributes." ) if options.endCol: endCol = int( options.endCol ) - 1 else: maf_utilities.tool_fail( "End column not set, click the pencil icon in the history item to set the metadata attributes." ) if options.strandCol: strandCol = int( options.strandCol ) - 1 else: strandCol = -1 if options.interval_file: interval_file = options.interval_file else: maf_utilities.tool_fail( "Input interval file has not been specified." ) if options.output_file: output_file = options.output_file else: maf_utilities.tool_fail( "Output file has not been specified." ) split_blocks_by_species = remove_all_gap_columns = False if options.split_blocks_by_species and options.split_blocks_by_species == 'split_blocks_by_species': split_blocks_by_species = True if options.remove_all_gap_columns and options.remove_all_gap_columns == 'remove_all_gap_columns': remove_all_gap_columns = True else: remove_all_gap_columns = True #Finish parsing command line #Open indexed access to MAFs if options.mafType: if options.indexLocation: index = maf_utilities.maf_index_by_uid( options.mafType, options.indexLocation ) else: index = maf_utilities.maf_index_by_uid( options.mafType, options.mafIndexFile ) if index is None: maf_utilities.tool_fail( "The MAF source specified (%s) appears to be invalid." % ( options.mafType ) ) elif options.mafFile: index, index_filename = maf_utilities.open_or_build_maf_index( options.mafFile, options.mafIndex, species = [dbkey] ) if index is None: maf_utilities.tool_fail( "Your MAF file appears to be malformed." ) else: maf_utilities.tool_fail( "Desired source MAF type has not been specified." ) #Create MAF writter out = bx.align.maf.Writer( open(output_file, "w") ) #Iterate over input regions num_blocks = 0 num_regions = None for num_regions, region in enumerate( bx.intervals.io.NiceReaderWrapper( open( interval_file, 'r' ), chrom_col = chromCol, start_col = startCol, end_col = endCol, strand_col = strandCol, fix_strand = True, return_header = False, return_comments = False ) ): src = maf_utilities.src_merge( dbkey, region.chrom ) for block in index.get_as_iterator( src, region.start, region.end ): if split_blocks_by_species: blocks = [ new_block for new_block in maf_utilities.iter_blocks_split_by_species( block ) if maf_utilities.component_overlaps_region( new_block.get_component_by_src_start( dbkey ), region ) ] else: blocks = [ block ] for block in blocks: block = maf_utilities.chop_block_by_region( block, src, region ) if block is not None: if species is not None: block = block.limit_to_species( species ) block = maf_utilities.orient_block_by_region( block, src, region ) if remove_all_gap_columns: block.remove_all_gap_columns() out.write( block ) num_blocks += 1 #Close output MAF out.close() #remove index file if created during run maf_utilities.remove_temp_index_file( index_filename ) if num_blocks: print "%i MAF blocks extracted for %i regions." % ( num_blocks, ( num_regions + 1 ) ) elif num_regions is not None: print "No MAF blocks could be extracted for %i regions." % ( num_regions + 1 ) else: print "No valid regions have been provided." if __name__ == "__main__": __main__()